A spectral approach for the exact observability of infinite dimensional systems with skew-adjoint generator. Karim Ramdani, Takeo Takahashi, Gérald Tenenbaum, Marius Tucsnak To cite this version: Karim Ramdani, Takeo Takahashi, Gérald Tenenbaum, Marius Tucsnak. A spectral approach for the exact observability of infinite dimensional systems with skew-adjoint generator.. Journal of Functional Analysis, Elsevier, 5, 6, pp.193-9. <hal-91371> HAL Id: hal-91371 https://hal.archives-ouvertes.fr/hal-91371 Submitted on 6 Sep 6 HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
A Spectral Approach for the Exact Observability of Infinite Dimensional Systems with Skew-Adjoint Generator K. Ramdani, T. Takahashi, G. Tenenbaum and M. Tucsnak Institut Elie Cartan Université Henri Poincaré Nancy 1, BP 39 Vandœuvre lès Nancy 5456, France Abstract: Let A be a possibly unbounded skew-adjoint operator on the Hilbert space X with compact resolvent. Let C be a bounded operator from D(A to another Hilbert space Y. We consider the system governed by the state equation ż(t = Az(t with the output y(t = Cz(t. We characterize the exact observability of this system only in terms of C and of the spectral elements of the operator A. The starting point in the proof of this result is a Hautus type test, recently obtained in Miller [19]. We then apply this result to various systems governed by partial differential equations with observation on the boundary of the domain. The Schrödinger equation, the Bernoulli-Euler plate equation and the wave equation in a square are considered. For the plate and Schrödinger equations, the main novelty brought in by our results is that we prove the exact boundary observability for an arbitrarily small observed part of the boundary. This is done by combining our spectral observability test to a theorem of Beurling on non harmonic Fourier series and to a new number theoretic result on shifted squares. Keywords: boundary exact observability, boundary exact controllability, Hautus test, Schrödinger equation equation, plate equation, wave equation. 1 Introduction and statement of the main results Let X be a Hilbert space endowed with the norm X, and let A : D(A X be a skew-adjoint operator. Assume that Y is another Hilbert space equipped with the norm Y and let C L(D(A, Y be an observation operator. According to Stone s theorem, A generates a strongly continuous group of isometries in X T = (T t t. This paper is concerned with infinite dimensional observation systems described by the equations ż(t = Az(t, z( = z, (1.1 1
y(t = Cz(t. (1. Here, a dot denotes differentiation with respect to the time t. The element z X is called the initial state, z(t is called the state at time t and y is the output function. Such systems are often used as models of vibrating systems (e.g., the wave equation, electromagnetic phenomena (Maxwell s equations or in quantum mechanics (Schrödinger s equation. In several particular cases we will also consider the control system which is the dual of (1.1, (1.. However, in order to avoid technicalities, we do not use the general form of the dual control system and we do not detail the duality arguments (we refer, for instance, to Tucsnak and Weiss [3] for a brief discussion of these issues. By a solution of (1.1 we mean that z(t = T t z (this is a mild solution. In order to give a sense to (1., we make the assumption that C is an admissible observation operator in the following sense (see Weiss [4]: Definition 1.1. The operator C in the system (1.1-(1. is an admissible observation operator if for every T > there exists a constant K T such that y(t Y dt K T z X z D(A. (1.3 If C is bounded, i.e. if it can be extended such that C L(X, Y, then C is clearly an admissible observation operator. Definition 1.. The system (1.1-(1. is exactly observable in time T if there exists k T > such that y(t Y dt k T z X z D(A. (1.4 The system (1.1-(1. is exactly observable if it is exactly observable in some time T >. The exact observability property is dual to the exact controllability property, as it has been shown in Dolecki and Russell [8]. By using the above duality, the exact controllability of a system governed by partial differential equations reduces to the observability estimate (1.4 (called inverse inequality in Lions [16]. Most of the literature tackling exact observability and exact controllability for systems governed by partial differential equations is based on a time domain approach. This means that one considers directly solutions of (1.1 (or of a dual equation which are manipulated in various ways: non harmonic Fourier series (Avdonin and Ivanov [] and references therein, multipliers method (Komornik [14], Lions [16] or microlocal analysis techniques (Bardos, Lebeau and Rauch [6]. Only few papers in the area of controllability and observability of systems governed by partial differential equations have considered a frequency domain approach, related to the classical Hautus test in the theory of finite dimensional systems (see Hautus [9]. Roughly speaking, a frequency domain test for the observability of
(1.1-(1. is formulated only in terms of the operators A, C and of a parameter (the frequency. This means that the time t does not appear in such a test and that we do not have to solve an evolution equation. In the case of a bounded observation operator C, such frequency domain methods have been proposed in Liu [17] and Liu, Liu and Rao [18]. In the case of an unbounded observation operator C a Hautus type test has been recently obtained in Miller [19]. The aim of this paper is to use Hautus type tests in order to characterize the exact observability property only in terms of C and of the spectral elements of the operator A. This will be done provided that the operator A has a compact resolvent and therefore, that the spectrum of A is formed only by eigenvalues. More precisely, since A is skew-adjoint, it follows that the spectrum of A is given by σ(a = {iµ n n Λ} with Λ = Z or Λ = N and where (µ n n Λ is a sequence of real numbers. The main result of this paper reads as follows: Theorem 1.3. Assume that A is skew-adjoint with compact resolvent and that the operator C is admissible for the system (1.1-(1.. Moreover, assume that (Φ n n Λ is an orthonormal sequence of eigenvectors of A associated to the eigenvalues (iµ n n Λ. For ω R and ε >, set J ε (ω = {m Λ such that µ m ω < ε}. (1.5 Then the system (1.1, (1. is exactly observable if and only if one of the following equivalent assertions holds : 1. There exists ε > and δ > such that for all ω R and for all z = c m Φ m : m J ε(ω Cz Y δ z X. (1.6. There exists ε > and δ > such that for all n Z and for all z = c m Φ m : m J ε(µ n Cz Y δ z X. (1.7 Remark 1.4. The above theorem can be seen as a generalization of several results in the literature. More precisely, in the particular case of a bounded observation operator C, the result in Theorem 1.3 follows, via a standard argument, from Theorem 3. in [18]. For unbounded C, but with the additional assumption that the sequence (µ n satisfies the gap condition (i.e., there exists γ > such that µ n µ m > γ for all m, n Λ, m n, the necessity of condition (1.6 in Theorem 1.3 is a consequence of Theorem 4.4 from Russell and Weiss [1]. 3
An important part of this paper is devoted to the application of the spectral criteria in Theorem 1.3 to systems governed by partial differential equations. The Schrödinger equation, the Bernoulli-Euler plate equation and the wave equation in a square are considered. For the plate and Schrödinger equations, the main novelty brought in by our results is that we show that the exact observability property can hold for an arbitrarily small observed part of the boundary. More precisely, in the case of the plate equation, our observability result implies the following exact controllability result. Theorem 1.5. Consider the square Ω = (, π (, π and let be an open subset of Ω. Consider the following control problem ẅ + w =, x Ω, t >, (1.8 w(x, t =, x Ω, t >, (1.9 w(x, t =, x Ω \, t > (1.1 w(x, t = u, x, t > (1.11 w(x, = w (x, ẇ(x, = w 1 (x, x Ω, (1.1 where the input is the function u L (, T; L (. Then the following assertions are equivalent: 1 For all T >, the above system is exactly controllable in H 1 (Ω H 1 (Ω in time T. This means that, for all (w, w 1 H 1 (Ω H 1 (Ω, we can find u L (, T; L ( such that w(x, T =, ẇ(x, T = x Ω. The control region contains both a horizontal and a vertical segment of non zero length. The proof of the above result is based on a consequence of Theorem 1.3 combined to a theorem of Beurling on non harmonic Fourier series and to a new number theoretic result (a theorem on shifted squares. Let us mention that in the case of a control acting in an arbitrary open subset of the square Ω, an exact controllability result for the plate equation has been given in Jaffard [11]. Moreover, we consider a system governed by the wave equation in a square. We give a very simple proof (it uses only Parseval s theorem of the boundary observability of this system. The paper is organized as follows. Section is devoted to the proof of our main result, namely Theorem 1.3. In Section 3, this result is applied to study the boundary observability of Shrödinger equation in a square. The case of a Dirichlet boundary observation and the Neumann one are successively considered. In Section 4, we derive the counterpart of Theorem 1.3 for second order systems (see Proposition 4.5. Thanks to this result, we tackle in Section 5 the problem of the boundary observability for the Bernoulli-Euler plate equation in a square. A second application of the spectral criteria provided by Proposition 4.5 is detailed in Section 6. This 4
application concerns the boundary observability of the wave equation in a square. Finally, Section 7 is devoted to the proof of Proposition 7.1, which is one of the main ingredients used to establish our observability results in Sections 3 and 5. Proof of Theorem 1.3 The basic tool in the proof of Theorem 1.3 is a recent Hautus type test. This result, given in [19], concerns the observability of systems with skew-adjoint generator and with unbounded observation operator. We first recall this result (see [19] for the proof. Theorem.1. Assume that A is a skew-adjoint operator in the Hilbert space X and that C : D(A Y is an admissible observation operator. Then the system (1.1, (1. is exactly observable if and only if there exists a constant δ > such that (A iωiz X + Cz Y δ z X ω R, z D(A. (.1 In order to prove Theorem 1.3, we need the following consequence of the admissibility property. Lemma.. Assume that the operators A and C are as in Theorem.1. Then there exists M > such that C(A I iωi 1 L(X,Y M, ω R. (. Proof. Our proof is a slight variation of the proof of Proposition.3 in [1]. Let us fix T > and z X. Then, for any n N we have that nt n 1 e t/ CT t z Y dt (k+1t e kt CT t z Y dt. (.3 k= By using the definition (1.3 of admissibility, combined to the fact that T is a group of isometries we have that (k+1t kt CT t z Y dt K T z X k N. The above inequality combined to (.3 implies that e t/ CT t z Y dt kt K T 1 e T z X z X. (.4 On the other hand C(A I iωi 1 z Y = e t CT t zdt 5 Y,
which, by applying the Cauchy-Schwartz inequality, yields : ( ( C(A I iωi 1 z Y e t dt e t/ CT t z Y dt. Combined to (.4, the above relation clearly implies the desired conclusion (., K T with M =. 1 e T We will also need the following result which can be seen as a generalization of Lemma 4.6 in [1]: Lemma.3. Assume that the operators A and C are as in Lemma.. For each ε > and ω R, we define the subspace V (ω X by V (ω = {Φ m m J ε (ω}, (.5 where J ε (ω is defined in (1.5. We denote by A ω the part of A in V (ω, i.e., A ω : D(A V (ω V (ω and A ω z = Az Then, there exists M > such that z D(A V (ω. C(A ω iωi 1 L(V (ω,y M, ω R. (.6 Proof. Given ω R, set s = 1+iω. Then, thanks to the resolvent identity, we have We first show that Indeed, let f = (A ω iωi 1 = (A ω si 1 [ I (A ω iωi 1]. (.7 m J ε(ω (A ω iωi 1 L(V (ω 1 ε. (.8 f m Φ m be an element of V (ω. Then (A ω iωi 1 f = m J ε(ω f m µ m ω. The above relation and the fact that µ m ω ε for m J ε (ω clearly imply (.8. On the other hand, we clearly have C(A ω si 1 L(V (ω,y C(A si 1 L(X,Y and thus, by using Lemma., we obtain that there exists a constant M > such that C(A ω si 1 L(V (ω,y M ω R. The above relation, (.7 and (.8 yield then (.6. 6
We are now in position to prove Theorem 1.3. Proof of Theorem 1.3. We first show that assertions 1 and in Theorem 1.3 are equivalent. It is clear that assertion 1 implies assertion (take ω = µ n. Conversely, assume that assertion holds true for some ε >, and let ω R. Then, either J ε/ (ω is empty, or there exists n J ε/ (ω and in this latter case, one can easily check that J ε/ (ω J ε (µ n. Consequently, in both cases, assertion 1 holds true. It remains to show that the exact observability of the system (1.1, (1. is equivalent to assertion 1. To achieve this, we use the characterization of exact observability provided by Theorem.1. Assume that the system (1.1, (1. is exactly observable. By Theorem.1, there exists a constant δ > such that (A iωiz X + Cz Y δ z X, (.9 for all ω R, and for all z D(A. On the other hand, for z = c m Φ m and for ε small enough, we have that m J ε(ω (A iωiz X = m J ε(ω By applying (.9 to z = 1 holds. m J ε(ω i(µ m ωc m ε z X δ z X. (.1 c m Φ m and by using (.1, we obtain that assertion Let us now assume that the system (1.1, (1. is not exactly observable. Then, by Theorem.1, condition (.1 is not satisfied, i.e., there exists sequences (ω n n N in R and (z n n N in D(A such that and satisfying z n X = 1 n N. (.11 lim (A iω nz n X = n lim Cz n Y =. (.1 n We introduce then the following orthogonal decomposition of z n = m Λ c n m Φ m: z n = z n + z n, (.13 with zn = c n mφ m, z n = c n mφ m, m J(ω n m/ J(ω n 7
where J(ω is defined for all ω R by relation (1.5. Let us prove that the sequences (ω n n N and (z n n N contradict assertion 1. First of all, we note that the orthogonality of (A iω n z n and (A iω n z n implies that (A iω n z n X (A iω n z n X = m/ J n (µ m ω n c n m ε z n X. The above relation and (.1 imply that and that lim z n X =, n Thanks to (.11 and (.13, the above relation yields On the other hand, (.13 implies that Moreover, using the notation of Lemma.3, we have lim (A iω n z n X = (.14 n lim n z n X = 1. (.15 Cz n Y Cz n Y + C z n Y. (.16 C z n = C(A ωn iω n 1 (A ωn iω n z n. Consequently, Lemma.3 implies that there exists M > such that C z n Y M (A ωn iω n z n X n N. The above relation and (.14 imply that lim C z n Y =. n This fact, together with (.1 and (.16 yield lim n Cz n Y =. The above relation and (.15 show that the sequences (ω n n N and (z n n N contradict assertion 1 in Theorem 1.3. 3 Boundary observability of the Schrödinger equation in a square 3.1 Dirichlet boundary observation Consider the square Ω = (, π (, π and let be an open subset of Ω. We consider the following initial and boundary value problem: ż + i z =, x Ω, t, (3.1 z =, ν x Ω, t, (3. z(x, = z (x, x Ω, (3.3 8
with the output y = z. (3.4 The system can be described by equations of the form (1.1, (1., if we introduce the appropriate spaces and operators. We first define the state space X = L (Ω and the operator A : D(A X by D(A = {ϕ H (Ω ϕ ν = }, (3.5 Aϕ = i ϕ ϕ D(A. (3.6 We next define the output space Y = L ( and the observation operator C L(D(A, Y Cϕ = ϕ ϕ D(A. (3.7 Proposition 3.1. With the above notation, C is an admissible observation operator. In other words, for all T there exists a constant K T > such that the if z, y satisfy (3.1-(3.4 then y ddt K T z L (Ω z D(A. We skip the proof of the above result since it can be easily obtained from the Fourier series expansion of the solution of (3.1-(3.3. The main result in this subsection is: Proposition 3.. For any non empty open subset of Ω, the system described by (3.1-(3.4 is exactly observable. In other words there exists T > and a constant k T > such that if z, y satisfy (3.1-(3.4 then z ddt k T z L (Ω z D(A. Proof. We have seen in Proposition 3.1 that C is an admissible observation operator for (3.1-(3.4 in the sense of Definition 1.1. On the other hand, A is clearly skewadjoint. Moreover, since the imbedding H 1 (Ω L (Ω is compact, A has a compact resolvent. Consequently, according to Theorem 1.3, it suffices to check that the operators A and C defined by (3.5, (3.6 and (3.7 satisfy condition in Theorem 1.3. The eigenvalues of A are µ m,n = i(m + n, m, n N. A corresponding orthonormal basis of X = L (Ω formed by eigenfunctions of A is Φ m,n (x 1, x = π cos (mx 1 cos (nx m, n N. 9
In order to check that condition in Theorem 1.3 holds, we have to show that there exists ε, δ > such that for all (q, r N N and for all z = c m,n Φ m,n, we have where Cz Y = (m,n J ε(µ q,r c m,n Φ m,n d δ (m,n J ε(µ q,r (m,n J ε(µ q,r J ε (µ q,r = {(m, n N N ; (m + n q r < ε}. It is clear that if we choose ε < 1 then J ε (µ q,r = {(m, n N N ; m + n = q + r }. c m,n, (3.8 Moreover, without loss of generality, we can assume that there exists α, β (, π with α < β and (α, β {}. Let S denote the set of squares of positive integers. For q, r N we set and for m Λ qr we put We have Λ qr = {m N q + r m S }, (3.9 f(m = q + r m. (3.1 Cz Y 4 β π c m,f(m cos(mx 1 α m Λ qr dx 1. (3.11 By using Proposition 7.1, the above relation implies that there exists a constant δ > such that 4 β π c m,f(m cos(mx 1 dx 1 δ c m,n. (3.1 α m Λ qr m +n =q +r From (3.11 and (3.1, we clearly get the desired estimate (3.8. By a standard duality argument, the above proposition implies that the following exact controllability holds. Corollary 3.3. For any non empty open subset of Ω, the system ż + i z =, x Ω, t >, z =, ν x Ω \, t >, z ν = u L (, T; L (, x, t >, z(x, = z (x, x Ω. is exactly controllable in some time T in the state space L (Ω. 1
3. Neumann boundary observation The example studied in this subsection differs from the case considered in the previous one only by the boundary condition and the choice of the observation operator. We prove that, in order to get exact observability we need a supplementary assumption on the observed part of the boundary. Consider the square Ω = (, π (, π and let be an open non empty subset of Ω. We consider the following initial and boundary value problem: ż + i z =, x Ω, t, (3.13 z =, x Ω, t, (3.14 z(x, = z (x, x Ω, (3.15 with the output y(t = z ν. (3.16 The system can be described by equations of the form (1.1, (1., if we introduce the appropriate spaces and operators. Indeed, let us first define the state space X = H 1 (Ω and the operator A : D(A X by D(A = {ϕ H 3 (Ω H 1 (Ω ϕ = on Ω}, (3.17 Aϕ = i ϕ ϕ D(A. (3.18 Next, we define the output space Y = L ( and the corresponding observation operator C L(D(A, Y by Cϕ = ϕ ν ϕ D(A. (3.19 Proposition 3.4. With the above notation, C is an admissible observation operator, i.e. for all T there exists a constant K T > such that if z, y satisfy (3.13-(3.16 then y ddt KT z L (Ω z D(A. The above result is classical (see, for instance, [15], so we skip its proof. The observability properties of the system (3.13-(3.16 are different from those encountered in the study of the system (3.1-(3.4. More precisely, if we denote by 1 = ([, π] {} ([, π] {π} the horizontal part of Ω and and by = ({} [, π] ({π} [, π] the vertical part of Ω, then the following result holds: Proposition 3.5. The system described by (3.13-(3.16 is exactly observable if and only if i, for i {1, }. In other words the following assertions are equivalent: 11
1. There exists T > and a constant k T > such that for all z, y satisfying (3.13-(3.16 we have y ddt k T z L (Ω z D(A.. The control region contains both a horizontal and a vertical segment of non zero length. Proof. We have seen in Proposition 3.1 that, for any open subset of Ω, C is an admissible observation operator for (3.1-(3.4 in the sense of Definition 1.1. Moreover, A is clearly skew-adjoint and it has compact resolvent. Therefore, we can apply condition in Theorem 1.3. The eigenvalues of A are µ m,n = i(m + n, m, n N. A corresponding family of normalized (in X = H 1 (Ω, eigenfunctions are Φ m,n (x 1, x = π m + n sin (mx 1 sin (nx, m, n N. We first show the necessity of condition i for i = 1,. Indeed, if this condition fails then we can assume, without loss of generality, that 1. We notice that Consequently, CΦ n,1 Y 1 Φ n,1 ν d = 8 1 π sin (nx π 1 + n 1 dx 1. (3. lim CΦ n,1 Y =, n which contradicts condition in Theorem 1.3. We next show that condition i for i = 1, implies that the operators A and C defined by (3.17, (3.18 and (3.19 satisfy condition in Theorem 1.3. In this case, without loss of generality we can assume that with < α i < β i < π, for i {1, }. For q, r N, we recall the notation ([α 1, β 1 ] {} ({} [α, β ], J ε (µ q,r = {(m, n N N ; (m + n q r < ε}. It is clear that if we choose ε < 1, then J ε (µ q,r = {(m, n N N ; m + n = q + r }. 1
If z = (m,n J ε(µ q,r Cz Y 4 π β1 α 1 c m,n Φ m,n then m Λ qr f(mc m,f(m m + f(m sin(mx 1 dx 1 β + f(nc f(n,n α f(n + n sin(nx n Λ qr dx, (3.1 where Λ qr and f are defined in (3.9 and in (3.1. On the other hand, by using Proposition 7.1, we obtain that there exists a constant δ > such that β1 f(mc m,f(m α 1 m n Λ qr + f (m sin(mx 1 dx 1 δ f (m f (m + m c m,f(m m Λ qr = δ n m + n c m,n, and β α n Λ qr f(nc f(n,n f (n + n sin(nx dx m +n =q +r δ f (n f (n + n c f(n,n n Λ qr = δ m m + n c m,n. m +n =q +r By taking the sum of the two above inequalities we get β1 f(mc m,f(m α 1 m + f (m sin(mx β 1 dx 1 + f(nc f(n,n α f (n + n sin(nx m Λ qr n Λ qr δ dx m +n =q +r c m,n. (3. By using (3.1 and (3. we obtain that Cz Y δ z X which concludes the proof. By a standard duality argument, the above proposition implies that the following exact controllability holds. Corollary 3.6. With the notation in Proposition 3.5, the system ż + i z =, x Ω, t >, z =, x Ω \, t >, z = u L (, T; L (, x, t >, z(x, = z (x, x Ω, 13
is exactly controllable in some time T > (in the state space X = H 1 (Ω if and only if i, for i {1, }. Remark 3.7. It can be shown, by using techniques similar to those in [14] and in [15], that the observability and the controllability results in this section hold for any T >. 4 Frequency domain tests for the exact observability of second order systems In this section we investigate an important particular case fitting in the framework of Theorem 1.3. This case is obtained by considering second order evolution equations occurring in the study of vibrating systems. More precisely, let H be a Hilbert space equipped with the norm and let A : D(A H be a self-adjoint, positive and boundedly invertible operator, with compact resolvent. Consider the initial value problem: ẅ(t + A w(t =, (4.1 w( = w, ẇ( = w 1, (4. which can be seen as a generic model for the free vibrations of elastic structures such as strings, beams, ( membranes, plates or three-dimensional elastic bodies. Moreover, let C L D(A 1, Y be an observation operator. We first show the equivalence of two conditions which will be used to define a concept of admissibility for observed systems described by second order differential equations. Proposition 4.1. With the above notation, the following conditions are equivalent: 1 For every T > there exists a constant K T such that the solutions w of (4.1, (4. satisfy ( C ẇ(t Y dt K T w + w D(A 1 1 (w, w 1 D(A D(A 1. (4.3 For every T > there exists a constant K T such that the solutions w of (4.1, (4. satisfy C w(t Y dt K T ( w + w 1 D(A1 w D(A 1, w 1 H, (4.4 where D(A 1 stands for the dual space of D(A 1 with respect to the pivot space H. Proof. We first show that assertion 1 implies assertion. If w D(A 1, w 1 H then the solution w of (4.1, (4. satisfies w C([, T]; D(A 1 C 1 ([, T], H. Define v(t = t w(sds A 1 w 1. 14
Clearly, we have v(t + A v(t =, v( = A 1 w 1 D(A, v( = w D(A 1. Since we supposed that assertion 1 holds true it follows that C w(s Y ds = C v(s Y ds ( K T A 1 w 1 + w D(A 1 H for all (w, w 1 D(A D(A 1. Since A is an isometry from D(A 1 onto D(A 1, the above inequality implies that assertion holds true. We still have to show that assertion implies assertion 1. First, assume that w D(A 3, w 1 D(A. Then, the solution w of the system (4.1, (4. satisfies w C 1 ([, T]; D(A C ([, T]; D(A 1. If we set v(t = ẇ(t then v C([, T], D(A C 1 ([, T], D(A 1 satisfies v(t + A v(t =, v( = w 1 D(A, v( = A w D(A 1. Since we supposed that assertion holds, we deduce that C ẇ(s Y ds = C v(s Y ds ( K T w 1 + A w [D(A 1 ] ( = K T w 1 + w, [D(A 1 ] for all w D(A 3, w 1 D(A. A density argument shows that assertion 1 holds. In the remaining part of this paper we consider systems of the form (4.1, (4. with one of the two following outputs: y = C w, (4.5 or y = C ẇ, (4.6 We are now in a position to give a definition of the admissibility for second order problems: 15
Definition 4.. C is an admissible observation operator for (4.1, (4. if it satisfies one of the equivalent conditions in Proposition 4.1. We next state the equivalence of two conditions which will be used in order to define a concept of exact observability for observed systems described by second order differential equations. Proposition 4.3. With the notation in Proposition 4.1, the following conditions are equivalent: 1 For every T > there exists a constant k T > such that the solutions w of (4.1, (4. satisfy C ẇ(t Y dt k T ( w + w D(A 1 1 (w, w 1 D(A D(A 1. (4.7 For every T > there exists a constant k T > such that the solutions w of (4.1, (4. satisfy C w(t Y dt k T ( w + w 1 D(A1 w D(A 1, w 1 H. (4.8 We skip the proof of the above result since it is completely similar to the proof of Proposition 4.1. Definition 4.4. The system described by (4.1, (4. and (4.5 is exactly observable in time T if it satisfies one of the equivalent conditions in Proposition 4.3. We can now state the main result of this section Proposition 4.5. Let A : D(A H be a self-adjoint, positive and boundedly invertible operator, with compact resolvent and let C L(D(A 1, Y be an admissible observation operator for (4.1-(4.. Let us denote by (λ n n N the increasing sequence formed by the eigenvalues of A 1 and by (φ n n N a corresponding sequence of eigenvectors, forming an orthonormal basis of H. For all ω > and all ε >, let us define the set I ε (ω = {m N such that λ m ω < ε}. (4.9 Then, the following propositions are equivalent: i The system (4.1 (4.5 is exactly observable. ii There exists a constant δ > such that ϕ D(A, ω > : (ω A ϕ + ω C ϕ Y δ ωϕ. (4.1 16
iii There exists ε > and δ > such that for all ω > and all ϕ = m I ε(ω c m φ m : C ϕ Y δ ϕ. (4.11 iv There exists ε > and δ > such that for all n N and all ϕ = m I ε(λ n c m φ m : C ϕ Y δ ϕ. (4.1 Remark 4.6. The fact that condition ii in the above proposition is equivalent to the exact observability can be seen as a generalization of Theorem 3.4 in [17], where a similar Hautus type result has been proved in the case of a bounded observation operator C. Proof of Proposition 4.5. It can be easily checked that the system (4.1 (4.5 can be written ( in the form (1.1- w(t (1. provided that we define the state of the system by z(t = and that we ẇ(t make the following choice of spaces and operators: ( X = D (A 1 I H, A =, C = ( C A. (4.13 The Hilbert space X is endowed here with the norm X defined by ( z X = A1 ϕ ϕ + ψ, z = X. ψ i ii: By Theorem.1 there exists a constant δ > such that (A iωz X + Cz Y δ z X ω R, z D(A. (4.14 ( ϕ Taking in the above relation z =, where ϕ D(A 1 iωϕ, we obtain that Cz Y = ωc ϕ Y, z X ωϕ while (A iωz X = (ω A ϕ. Therefore, (4.14 implies (4.1, and ii holds true. ii iii: Let ε < λ 1. Then it is easy to see that if ω < ε, then I ε(ω = and iii holds. If ω ε then for all ϕ = c m φ m, we have (ω A ϕ = m I ε(ω m I ε(ω ω λ m c m ε 17 m I ε(ω (ω + λ m c m 9ε ωϕ.
Consequently, for ε small enough and for ϕ = By applying condition ii to ϕ = m I ε(ω (ω A ϕ δ ωϕ. m I ε(ω c m φ m, we have c m φ m and by using the above equation, we obtain iii iii iv: This implication obviously holds (take ω = λ n. iv i: In order to prove this assertion we use Theorem 1.3. Suppose that iv holds true. Without loss of generality, we can assume that the constant ε in iv satisfies ε < λ 1. Let A and C be defined by (4.13. it can be easily checked that the eigenvalues of A are (iµ n n Z where { λn if n N µ n =, λ n if ( n N. If we set φ n = φ n, for all n N, then an orthonormal family (in X of eigenvectors (Φ n n Z of A is given by Φ n = 1 1 φ n iµ n, n Z. φ n In order to prove i it suffices, by Theorem 1.3, to show that there exists ε > and δ > such that for all n Z and for all 1 z = c m Φ m = 1 c m φ m iµ m m J ε(µ n (4.15 c m φ m we have m J ε(µ n Cz Y δ z X. m J ε(µ n Let us consider first the case where µ n > in (4.15. Then, µ n = λ n, and thus we have J ε (µ n = I ε (λ n (since ε < λ 1. Let us denote by ϕ the second component of z ϕ = 1 Then, it can be easily checked that m I ε(λ n Cz = C ϕ, c m φ m. and that z X = ϕ. 18
Consequently, by applying then iv to ϕ we get that Cz Y δ z X. The case µ n < can be treated similarly, and the proof is thus complete. 5 Boundary observability for the Bernoulli-Euler plate equation in a square Consider the square Ω = (, π (, π and let be an open subset of Ω. We consider the following initial and boundary value problem: with the output ẅ + w =, x Ω, t, (5.1 w(x, t = w(x, t =, x Ω, t, (5. w(x, = w (x, ẇ(x, = w 1 (x, x Ω, (5.3 y(t = ẇ ν. (5.4 The system (5.1-(5.4 can be written in the form (4.1, (4., (4.5. More precisely, we define H = H 1 (Ω, D(A = {ϕ H 5 (Ω H 1 (Ω ϕ = ϕ = on Ω}, Y = L (, A ϕ = ϕ ϕ D(A, With the above choice of spaces and operators, one can easily check that A is self-adjoint, positive, boundedly invertible and that D(A 1 = {ϕ H 3 (Ω H 1 (Ω ϕ = on Ω}. Moreover, the dual space of D(A 1 with respect to the pivot space H is D(A 1 = H 1 (Ω. The output operator corresponding to (5.4 is C ϕ = ϕ ν ϕ D(A 1. Proposition 5.1. With the above notation, C L(D(A 1, Y is an admissible observation operator, i.e. for all T there exists a constant K T > such that if w, y satisfy (5.1-(5.4 then y ddt K T for all (w, w 1 D(A D(A 1. ( w H 3 (Ω + w 1 H 1 (Ω 19
The above result is classical and for its proof we refer, for instance, to [16, p.87]. In order to state the observability properties of the system (5.1-(5.4, let us denote by 1 = ([, π] {} ([, π] {π} the horizontal part of Ω and by = ({} [, π] ({π} [, π] its vertical part. Then, the following result holds: Proposition 5.. The system described by (5.1-(5.4 is exactly observable if and only if i, for i {1, }. In other words, the following statements are equivalent 1. There exists T > and a constant k T > such that if z, y satisfy (5.1-(5.4 then ( y ddt kt w H 3 (Ω + w 1 H 1 (Ω (w, w 1 D(A D(A 1.. The control region contains both a horizontal and a vertical segment of non zero length. Proof. By Proposition 5.1, C is an admissible observation operator for the system described by (5.1-(5.4. Moreover, the imbedding D(A 1 H is clearly compact. Consequently, we can apply Proposition 4.5. The eigenvalues of A 1 are λ m,n = m + n m, n N. A corresponding family of normalized (in H = H 1 (Ω eigenfunctions are φ m,n (x = π m + n sin (mx 1sin (nx m, n N, x = (x 1, x Ω. We first show the necessity of condition i = for i = 1,. If this condition fails then we can assume, without loss of generality, that 1. We notice that C φ n,1 Y φ n,1 ν d = 8 1 π sin (nx π 1 + n 1 dx 1. (5.5 Consequently, lim C φ n,1 Y =, n which contradicts condition iii in Proposition 4.5. 1 In the remaining part of the proof, we show that if i, for i {1, }, then the operators A and C satisfy condition iv in Proposition 4.5. More precisely, we prove that for all ε (, 1 there exists δ > such that, for all q, r N and for all ϕ = a m,n φ m,n, we have (m,n I ε(λ q,r C ϕ Y = (m,n I ε(λ q,r a m,n φ m,n ν d δ (m,n I ε(λ q,r a m,n, (5.6
where I ε (λ q,r = {(m, n N N m + n = q + r }. Without loss of generality, we can assume that ([α 1, β 1 ] {} ({} [α, β ], with < α i < β i < π, for i {1, }. Then, we have C ϕ Y 4 π β1 α 1 m Λ qr f(ma m,f(m m + f (m sin(mx 1 dx 1 β + f(na f(n,n α f (n + n sin(nx n Λ qr dx, (5.7 where the set Λ qr and the function f are defined in (3.9 and (3.1. The desired inequality (5.6 follows now directly from (3. which was established in the proof of Proposition 3.5. We conclude this section by remarking that Theorem 1.5 stated in Section 1 follows directly from the results already proved in this section. More precisely, the fact that the system is exactly controllable is some time T > follows from Proposition 5. by a standard duality argument. Showing that T can be chosen arbitrarily small can be achieved by slightly adapting a classical argument (see for instance [14, p.81] or the Appendix written by Zuazua in [16]. 6 Boundary observability of the wave equation in a square In this section, we consider the problem of observability of the wave equation with Neumann boundary observation for the wave equation. This problem has been tackled by a large number of papers by using various methods (see for instance [16] and references therein. However, besides in the one dimensional case, no direct Fourier series based proof seems to exist in the literature. We give such a proof in the case where the space domain is a square. If we except the use of Proposition 4.5, the basic ingredients of the proof are very simple (we only need Parseval s theorem. Consider the square Ω = (, π (, π and let = ([, π] {} ({} [, π]. We consider the following initial and boundary value problem: ẅ w =, x Ω, t, (6.1 w =, x Ω, t, (6. w(x, = w (x, ẇ(x, = w 1 (x, x Ω, (6.3 with the output y = w ν. (6.4 1
The system (6.1-(6.4 can be written in the form (4.1-(4.5 if we introduce the following notation: H = H 1 (Ω, D(A = {ϕ H 3 (Ω H 1 (Ω ϕ = on Ω}, Y = L (, A ϕ = ϕ C ϕ = ϕ ν ϕ D(A, ϕ D(A. One can easily check that, with the above choice of the spaces and operators, we have that A is self-adjoint, positive and boundedly invertible and D(A 1 = H (Ω H 1 (Ω, D(A 1 = L (Ω. Proposition 6.1. With the above notation, C L(D(A 1, Y is an admissible observation operator, i.e. for all T there exists a constant K T > such that if w, y satisfy (6.1-(6.4 then y ddt K T for all (w, w 1 (H (Ω H 1 (Ω H1 (Ω. ( w H 1(Ω + w 1 L (Ω The above proposition is classical (see, for instance, Lions [16, p.44], so we skip the proof. The main result of this section is the following. Theorem 6.. The system described by (6.1-(6.4 is exactly observable. In other words, there exists T > and k T > such that y ddt k T for all (w, w 1 (H (Ω H 1 (Ω H1 (Ω. ( w H 1(Ω + w 1 L (Ω Proof. By Proposition 6.1, C is an admissible observation operator for the system described by (6.1-(6.4. Moreover, A is clearly self-adjoint, positive, and boundedly invertible, whereas the resolvent of A is clearly compact. Consequently, we can apply Proposition 4.5. The eigenvalues of A 1 are λ m,n = m + n m, n N. A corresponding family of normalized (in H = H 1 (Ω eigenfunctions are φ m,n (x = π m + n sin (mx 1sin (nx m, n N, x = (x 1, x Ω. (6.5
In the remaining part of the proof, we show that the operators A and C satisfy condition iii in Proposition 4.5. More precisely, we prove that there exists ε, δ > such that for all ω > and for all ϕ = a m,n φ m,n, we have where C ϕ Y = (m,n I ε(ω a m,n φ m,n ν (m,n I ε(ω d δ (m,n I ε(ω I ε (ω = {(m, n N N ; λ m,n ω < ε}. Let us introduce some notation. We first set K ε (ω = {m N n N with (m, n I ε (ω}. a m,n, (6.6 It is clear that if m K ε (ω then m < ω + ε. For m K ε (ω we introduce the set L(m defined by { L(m = {n N (m, n I ε (ω} = n N } m + n ω ε. (6.7 Then, we have L(m = {n N (ω ε m n } (ω + ε m (6.8 if m ω ε and L(m = {n N n } (ω + ε m, if ω ε < m < ω + ε. By using (6.5 and the above notation we get that C ϕ Y = 4 π π na mn sin(mx 1 m + n dx 1 n L(m + 4 π ma mn π sin(nx m + n m K ε(ω n K ε(ω m L(n dx. (6.9 We are going to prove that if ε (, 1 and if m K 1 ε(ω satisfies m < (ω +ε/, then the cardinal κ m of the set L(m defined in (6.7 satisfies κ m = 1. Assume that ε (, 1 and m K 1 ε(ω satisfies m < (ω + ε/. We first remark that (ω + ε/ ω ε. (6.1 Indeed, if the above inequality is not satisfied, then we get that ω < 9ε, and consequently, ω + ε < 1ε < 1. On the other hand, the fact that m K ε (ω implies that m < ω + ε < 1, which is a contradiction. 3
We have thus shown that (6.1 holds. Consequently, L(m satisfies (6.8, and its cardinal κ m is given by κ m (ω + ε m (ω ε m + 1 4ωε = (ω + ε m + + 1. (6.11 (ω ε m On the other hand, since m (ε + ω/, we have that and therefore, by (6.11, we obtain that (ω + ε m > ω, κ m 4 ε + 1. Since ε (, 1, the above relation implies that κ 1 m = 1 for all m K ε (ω such that m (ε + ω/. The unique element of L(m is then denoted by l m. This fact, combined to (6.9 and to the orthogonality of the family (sin(mx m 1 in L (, π, yields the existence of a constant δ > such that C ϕ Y l m a mlm + l n a lnn. π π l n + n m K ε(ω m (ω+ε/ m + l m n K ε(ω n (ω+ε/ The above relation and the fact that there exists C > such that for m (ε+ω/, we have l m C, m + l m implies the existence of δ > such that C ϕ Y δ m K ε(ω m (ω+ε/ a mlm + n K ε(ω n (ω+ε/ a lnn. (6.1 Using the fact that for all (m, n I ε (ω, we have either m ε + ω or n ε + ω, we obtain that (m,n I ε(ω a mn m K ε(ω m (ω+ε/ a mlm + n K ε(ω n (ω+ε/ a lnn. The desired inequality (6.6 follows then from the above relation, together with relation (6.1. 4
7 An Ingham-Beurling type result and a theorem on shifted squares The following result plays a central rôle in the proof of the observability results in sections 3 and 5. Proposition 7.1. For q, r N, we set Λ qr = {m N q + r m S }, where S denotes the set of squares of positive integers. Then, for any non empty interval I, there exists a constant δ >, depending only on I, such that the inequality I a n e inx n Λ qr holds for all sequence (a n l (C. dx δ n Λ qr a n, The main ingredients of the proof of the above result are a version of a famous theorem of Beurling [7] on non harmonic Fourier series, and a number theoretic theorem concerning shifted squares. Let us first state the version of Beurling s result given in Theorem 1.5 in Baiocchi, Komornik and Loreti [5]. For the proof, we refer to [5]. Theorem 7.. Let (λ n n Z be a strictly increasing sequence of real numbers such that λ n+1 λ n γ, n Z, for some γ >. Moreover, assume that exists γ γ and M N such that λ n+m λ n γm, n N. Then, for any interval I of length l(i > π, there exists δ > depending only on γ γ, γ, M and I, such that I a n e inx dx δ a n, n N n N holds for all sequence (a n l (C. Note that, due to a misprint, the condition l(i > π γ been written l(i > πγ in [5]. in the above theorem has 5
Remark 7.3. The result in Theorem 7. can be seen as a generalization of a classical inequality proved by Ingham in [1]. For other generalizations and related questions we refer to Avdonin and Moran [1], [3], Baiocchi, Komornik and Loreti [4], Jaffard, Tucsnak and Zuazua [1] and Kahane [13]. Next, we give the second main ingredient of the proof of Proposition 7.1. Theorem 7.4. For positive integers M, N, V, let Z = Z(M, N, V denote the set of those integers n such that M < n M +N and V n is a square. For a suitable positive absolute constant C 1, we have Z C N log(n, where Z denotes the cardinality of the set Z. For the sake of clarity, we postpone the proof of this theorem to the end of this section. A useful consequence of Theorem 7.4 is the following. Corollary 7.5. Let M N and λ < λ 1 < < λ M be M +1 consecutive elements of Λ qr. Then, we have M λ M λ C log (M. (7.1 where C is the constant appearing in Theorem 7.4. Proof. For N N, denote by U(N the cardinal number of the set By Theorem 7.4 we clearly have Consequently {j 1 λ j λ + N}. U(N C N log (N N N. M = U(λ M λ C (λ M λ log [(λ M λ ]. Now observe that, since C 1, (7.1 plainly holds if λ M λ > M. Otherwise we have M C (λ M λ log (M, so that (7.1 is still valid. We are now in position to prove Proposition 7.1. 6
Proof of Proposition 7.1. Take γ > π, where l(i denotes the length of the interval I. Since log(m 3 M for all M 1, Corollary 7.5 implies that if M > 9C 4 γ l(i and if λ < λ 1 < < λ M are M+1 consecutive elements of Λ qr, then λ M λ Mγ. Moreover, the distance between any two distinct elements of Λ qr is at least one. Therefore, we can apply Theorem 7. to get the desired inequality. In order to prove Theorem 7.4, we first introduce some notation. For any prime number p, let (Z/pZ be the (cyclic multiplicative group of invertible residues modulo p and let Q p denote the subset of (Z/pZ comprising all non zero quadratic residues. Recall that the Legendre symbol is the mapping from Z onto { 1,, 1} defined by the formula ( n := p 1 if n Q p (modp, if p n, 1 if n (Z/pZ \ Q p (modp, (n Z. A classical result states that, for all odd primes p and all integers n such that p n, we have ( n n (p 1/ (mod p. (7. p This will be used in the proof of the following lemma. The result is known see, for instance, [], Exercise 3.3. but, for convenience of the reader, we provide a short proof. Lemma 7.6. For any odd prime p and all a (Z/pZ, we have ( n + a = 1. p n<p Proof. Denote the sum on the left by S p (a. By (7., we have S p (a n<p (n + a (p 1/ n<p j (p 1/ j (p 1/ ( (p 1/ ( (p 1/ j n j a (p 1/ j j a (p 1/ j n j (modp. n<p Now observe that the inner sum is zero modulo p unless when j = (p 1/, in which case it is 1. This is a well-known consequence of the fact that (Z/pZ is cyclic and we omit the details. We thus obtain S p (a 1 (modp. 7
Since S p (a p, this leaves the two possibilities S p (a = p 1 and S p (a = 1. However the former case can only happen if exactly one of the Legendre symbols is while all others have value 1. If this holds, then we have, for some integer h [, p[, ( h + a =. p Since p a, we must have h. Thus h p h (modp and obviously ( (p h + a =, p a contradiction. Hence S p (a = 1, as required. We can now embark on the proof of Theorem 7.4. Proof of Theorem 7.4. Our initial strategy consists in showing that, for all primes p such that p 3 (mod4, the subset E p of Z/pZ comprising those residue classes which contain at least one element of Z is small in size. We consider two cases, according to whether p V or not. To deal with the first instance, we observe that ( 1 = ( 1 (p 1/ = 1, p so n is not a quadratic residue modulo p if p n. Thus V n can only be a square if it is divisible by p and in fact by p. Therefore, we have E p = 1 if p V. In the second case, we have ( V n n E p = 1 or. p Since there are at most two solutions of the equation V n (modp, we plainly derive E p 1 + 1 { ( V n } 1 + p n<p = 1 (p + + 1 ( 1 ( n V p p = 1 (p + 3, n<p where, in the last stage, we have appealed to Lemma 7.6. We have therefore shown that, for all primes p 3(mod4, the set Z is excluded from 1 (p 3 residue classes modulo p. By the large sieve (see e.g.[], Corollary I.4.6.1 this yields, for all Q 1, Z (N + Q /L (7.3 8
with L := q Q g(q where g(q := µ(q p q, p 3(mod 4 p 3 p + 3 q 1. Here, as usual in number theory, the letter p denotes a generic prime number and q µ(q denotes the Möbius function. It remains to evaluate L as a function of Q. To this end, we introduce the Dirichlet series associated to g, viz G(s := q 1 g(q q s = p 3(mod 4 ( 1 + p 3 (p + 3p s, where s is a complex parameter, with initially Res > 1. We need to express this quantity in terms of the Riemann zeta function ζ(s. This can be achieved by introducing the unique non principal character modulo 4, defined by χ(p = ( 1 (p 1/, and the corresponding L function L(s, χ := (n + 1 = s n p 3 ( 1 n ( 1 χ(p/p s 1. We have, still for Res > 1, G(s = ( {1 χ(p}(p 3 1 + (p + 3p s p 3 = p p 3(1 s 1/ (1 χ(pp s 1/ H(s = ζ(s 1/ L(s, χ 1/ (1 s 1/ H(s, with H(s := ( (1 p s 1/ (1 χ(pp s 1/ {1 χ(p}(p 3 1 + (p + 3p s p 3 ( 1 p s 1/( = 1 + p 3 1 + p s (p + 3p s p 3(mod 4 ( 1/ = (1 p s 1 6ps + p 3. (p + 3p s p 3(mod 4 Since L(s, χ 1/ has analytic continuation in the region σ 1 c/ log(3 + τ (s = σ + iτ for a suitable positive absolute constant c (see e.g.[], notes on sections II.8. and II.8.3 and since the product H(s converges for σ > 1 and is bounded in 9
any half-plane σ 1 + δ with δ >, we are in a position to apply Selberg Delange type estimates, as given in [], Theorem III.5.3. This yields g(q = q Q AQ ( 1 } {1 + O log Q log Q Q (7.4 with A := H(1 πl(1, χ = π p 3(mod 4 1/ ( (1 p 1 7p 3. p (p + 3 Inserting (7.4 into (7.3 and selecting Q := N furnishes the bound Z B { ( 1 } N log N 1 + O log N N, with B := A = π p 3(mod 4 ( 1/ (1 p 7p 3 1 + 5.3159. (p 1(p + 4p 3 This finishes the proof of Theorem 7.4. References [1] S. Avdonin and W. Moran, Ingham-type inequalities and Riesz bases of divided differences, Int. J. Appl. Math. Comput. Sci., 11 (1, pp. 83 8. Mathematical methods of optimization and control of large-scale systems (Ekaterinburg,. [] S.-A. Avdonin and S.-A. Ivanov, Families of exponentials - The method of moments in controllability problems for distributed parameter systems, Cambridge University Press, Cambridge, 1995. [3] S. A. Avdonin and W. Moran, Simultaneous control problems for systems of elastic strings and beams, Systems Control Lett., 44 (1, pp. 147 155. [4] C. Baiocchi, V. Komornik, and P. Loreti, Ingham type theorem and applications to control theory., Boll. Unione Mat. Ital. Sez. B, 8 (1999, pp. 33 63. [5] C. Baiocchi, V. Komornik, and P. Loreti, Ingham-Beurling type theorems with weakened gap conditions, Acta Math. Hungar., 97 (, pp. 55 95. [6] C. Bardos, G. Lebeau, and J. Rauch, Sharp sufficient conditions for the observation, control and stabilization of waves from the boundary, SIAM J. Control. and Optim., 3 (199, pp. 14 165. 3
[7] A. Beurling, The collected works of Arne Beurling. Vol., Contemporary Mathematicians, Birkhäuser Boston Inc., Boston, MA, 1989. Harmonic analysis, Edited by L. Carleson, P. Malliavin, J. Neuberger and J. Wermer. [8] S. Dolecki and D. L. Russell, A general theory of observation and control., SIAM J. Control Optimization, 15 (1977, pp. 185. [9] M. Hautus, Controllability and observability conditions of linear autonomous systems, Nederl. Akad. Wet., Proc., Ser. A, 7 (1969, pp. 443 448. [1] A. Ingham, Some trigonometrical inequalities with applications to the theory of series, Math. Zeischrift, 41 (1936, pp. 367 379. [11] S. Jaffard, Contrôle interne exact des vibrations d une plaque rectangulaire., Port. Math., 47 (199, pp. 43 49. [1] S. Jaffard, M. Tucsnak, and E. Zuazua, On a theorem of Ingham, J. Fourier Anal. Appl., 3 (1997, pp. 577 58. [13] J.-P. Kahane, Pseudo-périodicité et séries de Fourier lacunaires, Ann. Sci. École Norm. Sup. (3, 79 (196, pp. 93 15. [14] V. Komornik, Exact Controllability and Stabilization - The Multiplier Method, John Wiley and Masson, Chichester and Paris, 1994. [15] G. Lebeau, Contrôle de l équation de Schrödinger, J. Math. Pures Appl. (9, 71 (199, pp. 67 91. [16] J.-L. Lions, Contrôlabilité exacte, perturbations et stabilisation de systèmes distribués. Tome 1, vol. 8 of Recherches en Mathématiques Appliquées, Masson, Paris, 1988. [17] K. Liu, Locally distributed control and damping for the conservative systems, SIAM J. Control Optim., 35 (1997, pp. 1574 159. [18] K. Liu, Z. Liu, and B. Rao, Exponential stability of an abstract nondissipative linear system., SIAM J. Control Optimization, 4 (1, pp. 149 165. [19] L. Miller, Controllability cost of conservative systems: resolvent condition and transmutation, J. Func. Anal., to appear. [] I. Niven, H. S. Zuckerman, and H. L. Montgomery, An introduction to the theory of numbers.5th ed., John Wiley & Sons, New York, 1991. [1] D. L. Russell and G. Weiss, A general necessary condition for exact observability., SIAM J. Control Optimization, 3 (1994, pp. 1 3. [] G. Tenenbaum, Introduction to analytic and probabilistic number theory, vol. 46 of Cambridge Studies in Advanced Mathematics., Cambridge Univ. Press, Cambridge, 1995. 31
[3] M. Tucsnak and G. Weiss, Simultaneous exact controllability and some applications., SIAM J. Control Optimization, 38 (, pp. 148 147. [4] G. Weiss, Admissible observation operators for linear semigroups, Israel J. Math., 65 (1989, pp. 17 43. 3